Nvidia just announced the RTX Spark at Computex 2026 in Taipei, and Jensen Huang called it “the biggest PC reinvention in 40 years.” That’s a bold claim — but the specs back it up.
RTX Spark is a new superchip that combines a Blackwell GPU and a 20-core Grace CPU into a single chip with up to 128GB of unified memory and 1 petaflop of AI compute. The headline capability: running 120-billion-parameter LLMs locally, with 1 million tokens of context — all on a laptop with all-day battery life.
What this means for builders
Until now, running serious AI models locally meant a desktop with multiple GPUs or cloud API costs. RTX Spark changes that equation. You can run frontier-class models privately, offline, with no API fees and no data leaving your machine.
Nvidia and Microsoft are also shipping a new security layer called OpenShell that lets AI agents run safely on your primary PC — no sandboxed VM required. Adobe is rebuilding Photoshop and Premiere from scratch for it.
When and who
Laptops and compact desktops from ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI land this fall. No pricing yet, but this is clearly aimed at developers and creators, not just consumers.
If you’re building AI products that need on-device inference, this is the hardware inflection point you’ve been waiting for.